The goal of the proposed research is to determine how policies that affect geographic access to medical services and provider reimbursement influence treatment patterns, health care expenditures, and health outcomes for the elderly. It focuses on five common medical conditions that account for a substantial portion of mortality, morbidity, and medical expenditures in the elderly; acute myocardial infarction, chronic heart disease, cancer, stroke, and cataracts. The proposed research relies primarily on very large, longitudinal datasets that include information on all medical utilization covered by Medicare. For each clinical condition, the proposed research will: (1) Identify the effects of three factors unrelated to health status on clinical decision making and health care utilization patterns: patient distance to treatment facilities for the condition; the availability of intensive technologies to providers; and alternative provider payment policies. Methods will examine differences in these effects across sociodemographic groups. (2) Determine the consequences of variations in treatment patterns induced by these factors for the following patient outcomes: survival time, adverse health events reflected in subsequent utilization patterns, and overall costs of health care. The methods will provide unbiased estimates of the average effects of alternative medical treatments in 'marginal" patients, patients whose treatment is affected by the factors unrelated to health status. The methods, known as instrumental variables in econometrics, have rarely been applied in observational studies in biostatistics and epidemiology but may be widely applicable. Applications will also be explored in the estimation and interpretation of treatment effects over time in clinical trials with compliance or crossover problems. An important component of the proposed research is an analysis of the validity of the methods in large medical claims datasets, using a variety of statistical tests and supplementary datasets with much more clinical detail.